Upgraded REVIEW from a soft suggestion to a real state with real expectations. Before this week, REVIEW was basically "maybe someone should look at this." That is not a governance mechanism. That is a hope.
Now there is an actual approval trail with shape: who approved it, when, what decision context they had, and what rationale they provided. If a human approves an action, the system records it with the same rigor as an automated ALLOW. No second-class approvals.
Updated the receipt fields to represent review transitions cleanly. You can look at a receipt and see the full path: initial evaluation, routing to review, reviewer identity, approval timestamp, and the final execution that followed.
The key insight is that review needs to be fast and scoped. If reviewing an action takes 20 minutes of context gathering, teams will bypass it. The reviewer should see what the agent wants to do, what policy flagged it, and what the risks are. All in one screen. Anything that adds friction to review is actively making governance worse, not better.